Revisit the way the down sampling is performed in `xsar`

[agrouaze]

currently the default strategy of down-sampling is based on rasterio.enums.Resampling.rms.

• For Sentinel-1 the down-sampling is delegated to the reader in https://github.com/umr-lops/xarray-safe-s1/blob/4c0184c6a1580118ccb1d36062375ad2081acff4/safe_s1/reader.py#L249
• For RCM is is done in xsar : https://github.com/umr-lops/xsar/blob/3412f47ece0ec9670dbb01c4878dd88e2cd5b48d/src/xsar/rcm_dataset.py#L881
• For RadarSAT-2 it is done in the associated reader: https://github.com/umr-lops/xradarsat2/blob/de1141391ab0b197ade12e6427804c6e89b95d80/src/xradarsat2/radarSat2_tiff_reader.py#L125

1) Ideally the down-sampling should be done fully in xsar (I think with rasterio the change of resolution is done only at opening, see https://github.com/rasterio/rasterio/issues/755). The reader should only map the files to python objects, "editing" operations should be done in xsar. 2) Ideally the down-sampling should let the user choose:

• the posting of the "pools"
• the size of the "pools"
• the shape of the "pools"
• the overlap of the "pools"
• the operation to apply on the "pools"